Representation and optimization of software usage models with non-Markovian state transitions

We extend Markov usage models by admitting non-Markovian transitions between states. The suggested description syntax preserves the visualization capacities of Markov models and, although it is kept simple, provides the computational power of a universal programming language. It is shown that a previously introduced optimization technique for computing optimal rest transition probabilities can be generalized to the presented framework. A medium-size example from a real-world application illustrates the approach. (C) 2000 Elsevier Science B.V. All rights reserved.